Prediction capability of neural network models for higher heating value used for boiler efficiency

Abstract

Quantity of heat produced during the complete burning of a unit mass of coal is indicated as Higher Heating Value (HHV). Accurate measurement of HHV of coal within the specific time is an important step in mine processing and power plant operations for the optimized operation, emission control and economic operation of the coal fired power plant. HHV is computed using the coal properties which are obtained through the proximate analysis or ultimate analysis. The proximate analysis yields moisture, ash, volatile matter and fixed carbon whereas the ultimate analysis yields carbon, hydrogen, nitrogen, oxygen and Sulphur as coal properties. Existing techniques of HHV determination are either linear or non-linear. As the relationship between the HHV and Coal properties are non-linear in nature, Neural Network (NN) and Wavelet model are suitable for HHV determination. Compare to NN, wavelet model estimates more accurate results but does not support real-time and recursive characteristics seen in NN. Hybrid model of wavelet and NN called as Wavelet Neural Network (WNN) has good prediction capability in non-linear environment. Hence, the prediction capability of WNN with respect to HHV estimation is analyzed in this paper by comparing the Mean Square Error of WNN and NN. The obtained result from our developed WNN and NN model for HHV estimation shows that WNN is better than NN.